Swarm intelligence methods for feature selection and machine learning methods for multiple vehicle tracking and detection

Abstract

With the rapid development of intelligent video analysis, traffic newlinemonitoring has become a key technique for collecting information about newlinetraffic conditions. Thus, multiple vehicle tracking and detection plays a vital newlinerole in traffic monitoring. In the recent work, multiple vehicle tracking and newlinedetection methods produce indicate issues like the accurate localization of newlinetarget object in extreme conditions such as occlusion, scaling, illumination newlinechange, and shape transformation, all of which still remain a challenge due to newlineincorrect detection of edges, higher dimensional feature space, ghost newlineshadows, three-dimensional space, and detection. newlineThis has motivated us to introduce a new track towards multiple newlinevehicle tracking and detection methods in order to improve detection newlineefficiency and the tracking results. The process of multiple vehicle tracking newlineand detection involves the following tasks: (1) Feature Extraction, newline(2) Background and Foreground Segmentation, (3) Edge Detection, newline(4) Dimensionality Reduction, (4) Feature Selection, (5) Multiple Vehicles newlineDetection and Tracking, (6) Performance Evaluation. There are three major newlinecontributions made in this technical work for performing the abovementioned newlinesteps. newlineThe first contribution of the work - Enhanced Convolution Neural newlineNetwork with Support Vector Machine (ECNN-SVM) is introduced for newlinemultiple vehicle detection. In this work, Local Binary Pattern (LBP) and newlineConvolutional Neural Network (CNN) features are extracted from multiple newlinevehicles. Enhanced Bat Optimization (EBO) is introduced to select particular newlinefeatures from the extracted features. In the EBO, fuzzy membership function newlineis used to generate the random number. newline